If you’ve been sitting in planning meetings debating whether to standardise on Microsoft Copilot or Anthropic’s Claude, you can stop arguing. On 9th March, Microsoft made that question redundant.
They announced Copilot Cowork, and buried inside that announcement was something far bigger than a new feature. Microsoft is now officially a multi-model platform. Claude is in Copilot. The debate is over. You can use both.
Instead of asking Copilot a question and getting a response, you describe the outcome you want. Cowork turns that into a plan, executes it across your M365 apps (Outlook, Teams, Excel, SharePoint) and checks in with you as it goes. Think of it less like chatting with an AI, and more like delegating to a colleague who never sleeps and has read every email in your organisation.
It’s built in close collaboration with Anthropic. Microsoft has taken the same agentic technology that powers Claude Cowork (Anthropic’s desktop agent for Mac and Windows) and rebuilt it inside the M365 cloud, with all the enterprise security, governance, and data context that entails. Copilot Cowork is currently in Research Preview and will roll out more broadly through the Frontier programme in late March.
Unlock the power of Copilot with a readiness assessment
But Here’s The Bigger Story
The Cowork capability is genuinely interesting. But the more significant announcement is what it signals about Microsoft’s direction.
For the last few years, Copilot was essentially an OpenAI product inside Microsoft packaging. GPT models, Azure infrastructure, one vendor powering everything. Microsoft has invested around $14bn in OpenAI and holds roughly a 27% stake following OpenAI’s October 2025 restructuring, so this was never a casual relationship.
That’s no longer true. As of this week:
- Anthropic’s latest Claude models are available in Copilot Chat for Frontier programme users, not just in specialist tools, but in the main Copilot experience
- Anthropic models are available in Copilot Studio, so when you’re building agents, you can choose which model does what: OpenAI for one step, Claude for another, or anything from the Azure Model Catalogue
- Researcher, Copilot’s deep-dive reasoning agent, can now be powered by either OpenAI or Anthropic models. Your choice.
It’s worth remembering why this matters commercially. When Anthropic launched Claude Cowork in January, Microsoft’s stock dropped around 14% as investors questioned whether enterprise software incumbents could compete with what frontier AI labs were now offering directly. This announcement is in part Microsoft’s answer to that question.
“Your work is not limited by one brand of models.”
Microsoft
That’s not a throwaway marketing line. It’s a strategic position.
Satya Nadella has been saying this for a while.
Nadella’s view on models has been consistent if you’ve been paying attention: models are becoming commodities. The value isn’t in which model you use. It’s in how you ground it with your data, your workflows, and your business context.
He’s also talked about fungibility: the idea that Microsoft’s infrastructure should be able to run any model and not be locked into any single vendor. The software layer, the data layer, the governance layer. Those are the hard things. The model itself is increasingly interchangeable.
Copilot Cowork and the Anthropic integration aren’t a U-turn on OpenAI. They’re the first public proof that Microsoft meant what Nadella was saying. The platform is more important than any individual model running on it.
What this means in practice
You don’t have to choose between Copilot and Claude. If your organisation is M365-licensed, you’re getting Claude as an option within Copilot, not instead of it. Claude’s strengths in complex reasoning and multi-step tasks are now available inside the platform you’re already paying for, with M365’s security and compliance controls applied by default.
For Copilot Studio builders, you can now select Claude models when building agents, routing specific workloads to whichever model is best suited. That’s a practical tool for building better agents, not just flexibility for its own sake.
On licensing, Microsoft also announced M365 E7: The Frontier Suite this week, available from 1st May at approximately £74/user/month (based on the published US price of $99). It bundles M365 E5, M365 Copilot, and the new Agent 365 governance layer into a single SKU. If you’re already at E5 and paying separately for Copilot, it’s worth a conversation with your licensing partner about whether E7 makes sense. Note that Microsoft UK pricing may vary slightly from the converted figure.
If you’re still in the “should we do AI at all?” phase, this announcement is a useful one to share with stakeholders who’ve been worried about vendor lock-in or betting on the wrong horse. The horses are now in the same stable.
The bottom line
Microsoft has positioned Copilot as the place where enterprise AI happens, not as a single product, but as a platform that brings in the best of whatever’s available. The Anthropic partnership and Copilot Cowork are the clearest proof yet that they mean it.
Nadella’s bet is that the platform wins, not the model. Given what was announced this week, it’s a pretty good bet.
Want to talk through what this means for your Copilot or agent roadmap?